This post examines the work in Self-training For Few-shot Transfer Across Extreme Task Differences, accepted as an oral presentation at ICLR 2021. In this post, we break down the task of cross-domain few shot learning, present a bird’s eye overview of the techniques involved, and describe the proposed method - STARTUP. In the latter section of the post, we also attempt to extend the experiments to language. To the best of our knowledge, cross-domain few shot learning in a multilingual setting has not been explored before and it raises interesting questions. Our code is available at this link.
This post outlines a few more things you may need to know for creating and configuring your blog posts.